CraTer is a static program analysis tool which aims to extract features from buggy source code and stack trace when a Java program crashed. CraTer is also the prototype in paper Does the Fault Reside in a Stack Trace?. It first extracts features from the crash datasets, then uses these features to train a model to predict whether the crashing fault reside in the corresponding stack trace or not.
This project is developed on Java 8.0. The external libraries we used are WEKA and SPOON, which can be downloaded in the Maven repository Weka Stable and Spoon Core.
Currently, we have only five packages at present, i.e.,
sklse.yongfeng.analysis, which conducts static analysis on 3500 mutated crashes,
sklse.yongfeng.analysis, which conducts static analysis on real-world crashes,
sklse.yongfeng.data, which implements some data pre-processing work,
sklse.yongfeng.experiments, which contains the whole experiments processes,
and sklse.yongfeng.launcher, which provide the main entry of the project.
- Analysis.java analyzes the 3,500 mutated crash instances from 7 real-world projects.
- RepsUtilier.java splits the single crash from stack traces file and analyzes the features from the stack trace.
- CrashNode saves detailed information of each crash.
- MEDAnalyzer.java analyzes the method-level features.
- CLSAnalyzer.java analyzes the class-level features.
- SRCAnalyzer.java analyzes the project-level features.
- FeatureList.java describes the detail information of total 89 features we need to extract.
sklse.yongfeng.analysis.realcrash package conducts the experiments on real-world crashes, implementing the similar functions with the package sklse.yongfeng.analysis.
- FilesSearcher.java provide different file searchers for searching suitable datasets.
- RandomGenerator.java is used to randomly select/generate crashes from each project.
- InsMerge.java is used to merage individual datasets into one total dataset.
- StatisticalProject.java calculates the statistical information(distribution of inTrace and outTrace instances) of each dataset.
- Overall.java evaluates the prediction approach on the total datasets combined with 7 projects using cross validation.
- Single.java evaluates the prediction approach on each project using cross validation.
- ImbalanceProcessingAve.java conducts the contrast experiment between using imbalanced data processing methods and No strategy.
- FeatureSelectionAve.java conducts the contrast experiment between using feature selection methods and No strategy.
- FeatureRankingAve.java ranks the features by Pearson's Correlation, then output the top-10 feature list of each project.
- TopTenFeatureEvaluation.java evaluate the prediction approach on the each project using only top-10 features.
- FoldResultsAve.java is used to get each fold result in cross validation, and the results will be used in the Wilcoxon signed-rank test.
- Launcher.java provides the main entry of the project.
Beside the packages presented above, the CraTer.jar is the core part of the project, we can run it directly on terminate or CMD windows. What'more, libs folder includes all dependency jar files used in our project.files folder includes all dataset used in our experiments. resources/crashrep collects stack trace files of each project, resources/projs saves all source codes of each project, resources/projs_libs preserves jar formats of each project.
There are two main way to run the CraTer on your computer, and it's also a easy work to conduct secondary development on CraTer.
Import the CraTer project into Eclipse, there is almost a main function in every java file. Click run as Java application to get the experimental results. For example, if we set the arguments of run configuration to -help, we will see,
Use the CraTer.jar contained in the project. try to use it on terminate or CMD windows. The commands are quite simple, for example, try to type java -jar CraTer.jar -help on CMD windows to show the description and help information of the CraTer project.
